Heterogeneity in Customization of Recommender Systems By Users with Homogenous Preferences
summary
Recommender systems with user tuning show that people who are assigned the same preferences still configure algorithms very differently: studies run on Mechanical Turk use slider-based controls, a travel persona and an incentivized exercise task to compare behaviors. Results from 375 participants produce roughly about 30 clusters of distinct configuration patterns, and repeated trials (≈100 users × 10 profiles) yield a largest cluster only 40–50%, so designs must allow many alternative configurations that converge on the same recommendation.
AI-written summary · about
chapters (7)
- 0:00Start
- 1:32CONTROLLABILITY
- 2:56DESIGN HEURISTIC
- 4:52USER EXPERIMENTS
- 6:12TRAVEL RECOMMENDER
- 7:53EXERCISE RECOMMENDER
- 10:30THANK YOU
related talks
41mRecommender Systems: Crafting Effective Personalised Experiences
05mTowards Guidelines for Designing Human-in-the-Loop Machine Training Interfaces
43mRecommendation Systems in Market Research
42mRethinking AI Alignment and Evaluation under Heterogeneous Preferences
24mWhen Recommendations Systems Go Bad
58mRecommender Systems – A Walkthrough
29mRecommender Systems for Personalized Financial Advice
37mFood is personal! Design and impact of personalized recommendations in e-grocery